Proposing Algorithm to Localize and Extract Facial Information Using FaceNet and MTCNN

Facial recognition systems from photos or videos are widely applied in daily life such as surveillance systems, access management, and security investigations. It can be compared with other forms of biometrics such as fingerprint or iris recognition systems. Among the facial recognition systems developed so far, Facenet introduced by Google brings a high accuracy rate. Therefore, in this paper, we will build a face recognition system using MTCNN| to detect faces and extract features using Facenet and SVM to classify and recognize faces. The proposed algorithm consists of 5 steps, namely, preparing data for training; face detection from data with MTCNN; extracting the features of each face with the FaceNet Keras model; classifying feature vectors by SVM, and performing face recognition. The algorithm results in an accuracy of up to 99.63% with data from only 5 to 15 images for a person. This proves that the algorithm is feasible when applied in practice.